An inter-mesh deployment pipeline
My experience working with large-scale environment drift and what I invented to try to remedy the problems.
Mark Veidemanis
Hi. I sometimes write things online.
Latest posts
A box around a search button
Maybe all apps need annotation tools. Here's how I made mine, and how I work with it when finding random bugs vs working on plans.
When do your good ideas happen?
Most idea tools track what you decided. On the mesh, every idea keeps its capture time and later earns an honest outcome grade in the retrospective — so it can show which times of day your good ideas actually strike. The full loop, and the signal that falls out the end.
Storm: turn conversations into plans your agent can run
I have a lot of conversations that should turn into something and then don't. Storm reaches out from a utility account, gathers the replies into a capture, and folds captures into plans. On the mesh, a plan is something you can actually run.
Intelligent backtesting and optimisation for cryptocurrency trading
Every trading tool will show you a backtest that made money. Here is how Alpha and Fisk try to tell the truth instead: a cross-symbol rule engine, a backtester that runs the exact code it trades with, and walk-forward optimisation that flags curve-fitting out loud.
Candles and widgets: a reusable architecture
The best part about making composable architecture is that progress is exponential. We started with graphs, moved on to dashboards, then created a window manager, allowing all the apps to be used from a single page application. This wasn't a big refactor, this is how it is meant
AI-assisted Android integration with emulated devices
I got tired of manually pairing WhatsApp during connector development, so I built an Android emulator workflow where AI agents can scan QR codes, control apps, and make integration testing someone else’s problem.
Your money, your body, and your time, on one mesh
Three new Zeitgeist Mesh services — bank aggregation, Garmin biometrics and time blocking. The interesting part isn't any one of them; it's the seams between them.
Open for alpha testers
Now open for alpha testers! (also includes screenshots)
Make it impossible to do the wrong thing
A field note on converging a mesh of small services onto one durable quota ledger — and why the goal was never "add a feature" but "delete the ways to get it wrong."
Compare: a tiny tool for finding the actual cheapest product
I built a small tool called Compare because I keep running into the same annoying problem whenever I buy supplements, nootropics, skincare, batteries, groceries, or anything else sold in awkward quantities. The cheapest product on the page is often not the cheapest product in real life. One shop gives you
Repeatable builds from messy experiments
I was not trying to build an AI DevOps workflow. I was trying to install CachyOS on a ThinkPad . That is a different problem, but it has the same shape as real infrastructure work: destructive commands, half-remembered decisions, SSH sessions, disk layouts, bootloader choices, and the quiet terror of realising
Setting up CachyOS on a ThinkPad T14s Gen 1 AMD with LUKS2, BTRFS, Limine and Sway
I wanted a boring laptop. Not boring in the “old corporate machine with a dying battery” sense, although this is a ThinkPad, so there is always a bit of that. I mean boring in the useful sense: encrypted, recoverable, fast enough, boring to boot, boring to update, boring
Why switching between Signal and WhatsApp is operational debt
Signal has one bit of context, WhatsApp has another, Instagram has the lead, XMPP has the weird but important technical person, and somehow you are meant to be the database.
Fast AI is a product feature, not a model choice
The provider is not the product. The latency budget is.